Files
gstack/test/skill-e2e-office-hours-brain-writeback.test.ts
T
f30b7b788a v1.91.13.0 fix: severe bug wave (autoplan guard, land-and-deploy CI gate, project identity, gbrain, review gates) (#3013)
* fix(land-and-deploy): never merge over red, pending or missing CI (#2995)

/land-and-deploy gated only on required checks, so a repo with none
(every GitHub Free private repo, any repo without branch protection)
skipped the CI wait and could merge over red CI. The verdict now comes
from bin/gstack-ci-gate, a tested helper bound to the head SHA: all
checks by default, cancel/unknown buckets fail, NO_CHECKS needs explicit
per-head approval, and both merge commands (including --auto) run only
after a PASS for the same SHA they pin with --match-head-commit.

Co-authored-by: harshitgavita-07 <harshit.gavita@gmail.com>

* fix(slug): give nested-group projects their own state bucket and verify checkpoint identity (#3003, #2901)

gstack-slug kept only the last two remote path segments, so
customer-a/product/repo and customer-b/product/repo shared one bucket
and /context-restore could present another project's checkpoint.

- One canonical-remote rule (bash + TS, golden vectors) used by all six
  resolvers: lowercased host, no userinfo/port/.git, Azure DevOps and
  Bitbucket Server normalized. 3+ segment remotes get
  <last-two>-<16 hex>; two-segment remotes (all of GitHub) are unchanged.
- Versioned sticky cache (v2:) keeps cache hits git-free.
- context-save stamps the credential-free remote and root;
  context-restore only treats this project's checkpoints as latest,
  reports foreign/unverified ones first, and treats a different root
  with the same remote (Conductor handoff) as information.
- gstack-slug --adopt-legacy copies only files the user confirms.
- question-preference-hook reads the same versioned cache, so never-ask
  preferences fire; unknown identity asks instead of auto-deciding.

* fix(gbrain): brain context loads, managed sync works, memory sync stops duplicating (#2883, #2985, #2922)

- brain-context-load called 'gbrain list_pages' (an MCP tool name), so
  every skill's brain context was 0 bytes. It now uses 'gbrain list'
  with gbrain's real filter flags, and prints one line when queries fail.
- The code-stage sync always passes --no-pull (managed gbrain >= 0.51
  refuses without it); a gbrain that rejects the flag fails with an
  upgrade message and is never retried without it. stderr is streamed
  with a bounded tail and quoted in failure summaries.
- /sync-gbrain gains --sources (and GSTACK_MEMORY_INGEST_SOURCES); by
  default it skips types already synced through a federated curated
  source, so curated pages stop being re-duplicated.

Co-authored-by: Vitor Cepeda Lopes <vitor@vitorcepedalopes.com>
Co-authored-by: harshitgavita-07 <harshit.gavita@gmail.com>

* fix(codex,cso): clean Codex reviews stop failing; semgrep and trivy SARIF results import (#2769, #3011)

- Codex review gate is PASS / FAIL / UNVERIFIED. Untagged output (what
  a clean 'codex review --base' prints) is UNVERIFIED, not FAIL, and
  never a false PASS; P2/P3-only output passes.
- import-sarif resolves semgrep's undeclared %SRCROOT% and trivy's
  ROOTPATH (including git-URL targets) through explicit, captured
  mappings; declared bases must sit inside the source root or the
  run's checkout, URIs are percent-decoded, and rejected results are
  counted with rule, base and path instead of silently dropped.

* fix: version strings are not IPs, failed update checks are not 'up to date', private security reporting (#2856, #2786, #2984)

- The pre-push redact hook exempts a root VERSION file that is only a
  version string and the version token in CHANGELOG release headings;
  the engine allowlist is untouched.
- gstack-update-check caches an unreadable remote as CHECK_FAILED with
  a short TTL instead of UP_TO_DATE; /gstack-upgrade never reports
  'latest' after a failed check.
- SECURITY.md points reporters at GitHub private vulnerability
  reporting.
- pr-title-sync (pull_request_target, write token) pins actions/checkout
  to a reviewed SHA.

* fix(autoplan): publication guard accepts real Claude Code journals (#3007, #3009, #2968, #2977, #2986, #3006)

The guard required a session journal to start with a user message, but
Claude Code starts it with SessionStart hook attachments (and a forked
compacted session starts with a compact_boundary), so every phase was
denied for anyone with a SessionStart hook. Windows path spellings
(C:/, /c/, drive case) were rejected too.

- Reader accepts the captured startup, compact, clear and fork shapes
  (claude 2.1.284 fixture), computes root uniqueness over every
  candidate, and returns typed reason codes instead of one string.
- Hard denials (competing root, foreign cwd, sidechain, agent, cycle,
  malformed, identity) say what happened and how to run the reviews by
  hand; unflushed journals still retry.
- A well-formed, stable journal whose root shape is unrecognized now
  warns ('phase publication was NOT verified') and continues, without
  overriding Claude Code's own permission check; no bypass switch.
- Windows paths fold before validation, so '..' is still rejected.
- Periodic drift canary runs the reader on fresh journals from the
  pinned claude.

Co-authored-by: Charles M Berg <77517285+CharlesBerg85@users.noreply.github.com>
Co-authored-by: laurenbdaniels-coder <264343335+laurenbdaniels-coder@users.noreply.github.com>
Co-authored-by: Jordan Mayer <176445995+maydaycyber@users.noreply.github.com>
Co-authored-by: elmesery-dev <elmesery.dev@gmail.com>

* v1.91.13.0: severe bug fix wave release notes, stale gbrain verbs in docs

* test: seed the versioned slug cache in the memory-injection deny case (#2901)

The preference hook now resolves the project bucket from gstack-slug's
v2 slug-cache entry instead of basename(cwd); this fixture still wrote
preferences under the basename bucket without the entry the skill
preamble always writes first.

* fix: full-suite follow-ups for the fix wave

- gstack-slug's --help heredoc is over 512B, so it gets the BASH_COMPAT=50
  pipe-deadlock guard like the other bins.
- Refresh the codex/factory ship goldens for the plan-gates slug change.
- The periodic lane gains the autoplan journal drift canary (107 paid
  files); raise the eval-slices cap to 25 so every slice still starts
  at once until its recorded duration lets the planner pack it.

* fix(plan-eng-review): clearer Step 0 sequencing, write routes and section outputs

The workflow judge scored plan-eng-review's instructions at
actionability 3.67 (bar 4) on text unchanged from main. Resolve the
ambiguities it named together, without changing any gate or output:
Step 0 names the Read's order and keeps one STOP; the pending-call,
read-only, forbidden-write and failed-save rules become one Write
routes block; header/label limits and grid checks are concrete;
Performance gets a checklist; each section has a stated output shape;
Lake Score is defined. Same total length (parity cap).

---------

Co-authored-by: harshitgavita-07 <harshit.gavita@gmail.com>
Co-authored-by: Vitor Cepeda Lopes <vitor@vitorcepedalopes.com>
Co-authored-by: Charles M Berg <77517285+CharlesBerg85@users.noreply.github.com>
Co-authored-by: laurenbdaniels-coder <264343335+laurenbdaniels-coder@users.noreply.github.com>
Co-authored-by: Jordan Mayer <176445995+maydaycyber@users.noreply.github.com>
Co-authored-by: elmesery-dev <elmesery.dev@gmail.com>
2026-10-02 11:11:00 -07:00

337 lines
14 KiB
TypeScript

/**
* E2E: /office-hours brain-writeback path under fake gbrain CLI.
*
* The matched-pair check for v1.50.0.0's "brain-aware planning actually
* works under Claude Code" headline: prove that when a user runs
* /office-hours with gbrain on PATH, the agent actually calls
* `gbrain put office-hours/<slug>` with valid frontmatter.
*
* Approach:
* 1. Render with --respect-detection into a temporary --out-dir against
* a temp GSTACK_HOME that has detected:true. Point section links at
* the final fixture workdir with --link-root; never rewrite the repo.
* 2. Copy the rendered office-hours skill and all its carved sections
* into the temp workdir.
* Also write docs/gbrain-write-surfaces.md so the agent can read the
* template on demand (the compact block points to it).
* 3. Write a fake `gbrain` shell script into workdir/bin/ with robust
* argv quoting (printf %q) so heredoc payloads in --content survive
* shell-to-shell. The fake logs every invocation + writes payloads
* to a per-slug file for inspection.
* 4. Run /office-hours via runSkillTest with workdir/bin/ first on PATH.
* Feed a deterministic founder pitch + auto-decide instructions.
* 5. Assert the argv log contains `gbrain put office-hours/<slug>`, the
* payload file exists with valid YAML frontmatter, and entity stubs
* were created.
*
* Periodic tier (~$0.50-1/run via claude -p, matches nearby
* setup-gbrain-path4-* tests at touchfiles.ts:496-498).
*
* NOT verified by this test (out of scope, owned by docs/gbrain-write-surfaces.md):
* - That gbrain itself persists what `gbrain put` is told (gbrain's
* own contract)
* - That `.gbrain-source` doesn't re-route writes (gbrain's contract)
* - Source-targeting (no way to fake source resolution in a stub CLI)
*/
import { expect, beforeAll, afterAll } from 'bun:test';
import { resolveEvalModel } from '../lib/eval-model';
import { CAPTURE_LONG_MS } from './helpers/eval-budgets';
import { execFileSync, spawnSync } from 'child_process';
import {
chmodSync,
copyFileSync,
cpSync,
existsSync,
mkdirSync,
mkdtempSync,
readFileSync,
readdirSync,
rmSync,
writeFileSync,
} from 'fs';
import { tmpdir } from 'os';
import { join } from 'path';
import { runSkillTest } from './helpers/session-runner';
import { seedHermeticGstackHome } from './helpers/hermetic-env';
import {
ROOT,
runId,
describeIfSelected,
testConcurrentIfSelected,
logCost,
evalsEnabled,
finalizeEvalCollector,
} from './helpers/e2e-helpers';
import { EvalCollector } from './helpers/eval-store';
import { OFFICE_HOURS_BUN_GRACE_MS, runRecordedOfficeHoursAttempt } from './helpers/office-hours-attempt';
const evalCollector = evalsEnabled ? new EvalCollector('e2e', undefined, 'office-hours-brain-writeback') : null;
afterAll(() => finalizeEvalCollector(evalCollector));
describeIfSelected(
'Office Hours Brain Writeback E2E',
['office-hours-brain-writeback'],
() => {
let workDir: string;
let callsLogPath: string;
let payloadDir: string;
let childEnv: Record<string, string>;
beforeAll(() => {
workDir = mkdtempSync(join(tmpdir(), 'skill-e2e-brain-writeback-'));
const run = (cmd: string, args: string[]) =>
spawnSync(cmd, args, { cwd: workDir, stdio: 'pipe', timeout: 5000 });
run('git', ['init', '-b', 'main']);
run('git', ['config', 'user.email', 'test@test.com']);
run('git', ['config', 'user.name', 'Test']);
// Copy the founder pitch fixture into the workdir.
const briefSrc = join(
ROOT,
'test',
'fixtures',
'office-hours-brain-writeback',
'brief.md',
);
copyFileSync(briefSrc, join(workDir, 'pitch.md'));
// Generate the brain-aware variant outside the repository. The renderer
// writes every Claude skill, so snapshotting just office-hours cannot
// protect the rest of the source tree from a detection-enabled render.
const tmpHome = mkdtempSync(join(tmpdir(), 'gbrain-detect-home-'));
const renderDir = join(tmpHome, 'render');
writeFileSync(
join(tmpHome, 'gbrain-detection.json'),
JSON.stringify({
gbrain_local_status: 'ok',
gbrain_on_path: true,
gbrain_version: 'test-0.41.0',
}),
);
// office-hours is carved (v2 plan T9): GBRAIN_SAVE_RESULTS moved into
// sections/design-and-handoff.md. Check the UNION for the gbrain put
// block, then copy all sections so every rendered section link resolves.
const skillPath = join(renderDir, 'office-hours', 'SKILL.md');
const sectionPath = join(renderDir, 'office-hours', 'sections', 'design-and-handoff.md');
try {
execFileSync(
'bun',
[
'run',
'scripts/gen-skill-docs.ts',
'--host',
'claude',
'--respect-detection',
'--out-dir',
renderDir,
'--link-root',
workDir,
],
{
// Read repository templates; all generated output lands in renderDir.
cwd: ROOT,
env: { ...process.env, GSTACK_HOME: tmpHome },
stdio: ['ignore', 'pipe', 'pipe'],
timeout: 60_000,
},
);
const brainAwareSkill = readFileSync(skillPath, 'utf-8');
const hasSection = existsSync(sectionPath);
const brainAwareSection = hasSection ? readFileSync(sectionPath, 'utf-8') : '';
if (!(brainAwareSkill + brainAwareSection).includes('gbrain put "office-hours/')) {
throw new Error(
'Regenerated office-hours skeleton+section does not contain gbrain put block. ' +
'Detection override may be broken — see test/gbrain-detection-override.test.ts.',
);
}
cpSync(join(renderDir, 'office-hours'), join(workDir, 'office-hours'), { recursive: true });
} finally {
rmSync(tmpHome, { recursive: true, force: true });
}
// Copy docs/gbrain-write-surfaces.md so the compact resolver block's
// on-demand reference resolves (the agent may read it for the full
// template; we don't require this read but make it available).
const docsSrc = join(ROOT, 'docs', 'gbrain-write-surfaces.md');
const docsDst = join(workDir, 'docs', 'gbrain-write-surfaces.md');
mkdirSync(join(workDir, 'docs'), { recursive: true });
copyFileSync(docsSrc, docsDst);
// Set up the fake gbrain CLI with robust argv quoting + payload capture.
callsLogPath = join(workDir, 'gbrain-calls.log');
payloadDir = join(workDir, 'gbrain-payloads');
mkdirSync(payloadDir, { recursive: true });
const binDir = join(workDir, 'bin');
mkdirSync(binDir, { recursive: true });
const fakeGbrain = `#!/bin/bash
# Fake gbrain CLI for E2E test. Logs every invocation with shell-safe quoting
# (printf %q) so --content "$(cat <<'EOF' ... EOF)" payloads survive intact.
{ printf 'gbrain'; for a in "$@"; do printf ' %q' "$a"; done; printf '\\n'; } \\
>> "${callsLogPath}"
case "$1" in
--version) echo "gbrain test-0.41.0"; exit 0 ;;
search) echo "[]"; exit 0 ;;
# Real gbrain: \`get <slug>\` reads a page back; \`get_page\` is an MCP tool
# name, not a CLI verb.
get) cat "${payloadDir}/$2.md" 2>/dev/null; exit 0 ;;
get_page) echo "Unknown command: get_page" >&2; exit 1 ;;
put)
SLUG="$2"
shift 2
while [ -n "$1" ]; do
if [ "$1" = "--content" ]; then
PAYLOAD_DIR="${payloadDir}"
mkdir -p "$PAYLOAD_DIR/$(dirname "$SLUG")"
printf '%s' "$2" > "$PAYLOAD_DIR/$SLUG.md"
break
fi
shift
done
exit 0
;;
esac
exit 0
`;
const fakePath = join(binDir, 'gbrain');
writeFileSync(fakePath, fakeGbrain);
chmodSync(fakePath, 0o755);
// Keep runtime state inside this fixture. Reuse the standard seed so
// isolating HOME does not introduce onboarding into the writeback case.
const fixtureHome = join(workDir, '.fixture-home');
const gstackHome = join(fixtureHome, '.gstack');
mkdirSync(join(fixtureHome, '.gbrain'), { recursive: true });
mkdirSync(gstackHome, { recursive: true });
seedHermeticGstackHome(gstackHome);
childEnv = {
HOME: fixtureHome,
GBRAIN_HOME: fixtureHome,
GSTACK_HOME: gstackHome,
PATH: `${binDir}:${process.env.PATH || ''}`,
};
run('git', ['add', '.']);
run('git', ['commit', '-m', 'fixture']);
});
afterAll(() => {
try {
rmSync(workDir, { recursive: true, force: true });
} catch {
// best effort
}
});
testConcurrentIfSelected(
'office-hours-brain-writeback',
async () => {
await runRecordedOfficeHoursAttempt({
collector: evalCollector,
name: '/office-hours-brain-writeback',
suite: 'Office Hours Brain Writeback E2E',
model: resolveEvalModel('capture'),
run: (signal) => runSkillTest({
signal,
prompt: `Read office-hours/SKILL.md for the workflow.
Read pitch.md — that's a founder pitch coming to office hours. Select Startup Mode. Skip any AskUserQuestion — this is non-interactive; auto-decide the recommended option for any question.
For the diagnostic, assume the founder confirmed Q1 (strongest evidence = "230 from a single tweet + 51 paying creators in 6 weeks"), Q2 (status quo = "creators write ad-hoc checks or use opaque Patreon-style platforms"), and Q3 (forcing question already asked).
Generate the design doc per Phase 5. The feature-slug value to substitute into the SAVE_RESULTS template's \`<feature-slug>\` placeholder is exactly 'pixel-fund' (no path prefix — the template already provides the prefix). The \`gbrain\` binary is on PATH at ${workDir}/bin/gbrain. Apply the SAVE_RESULTS template literally: the slug should land at \`<prefix>/pixel-fund\` per the resolver shape, with the actual design doc markdown body in the --content payload. Then enrich entity stubs for any named people or companies mentioned in the pitch.
This is a test of the brain-writeback path. Do NOT skip the gbrain save step under any circumstance — the runtime guard ("skip if gbrain not on PATH") does NOT apply here because gbrain IS available. Do NOT explore gbrain --help; follow the SAVE_RESULTS template's exact CLI shape. If you encounter any AskUserQuestion, auto-decide recommended.`,
workingDirectory: workDir,
maxTurns: 12,
timeout: CAPTURE_LONG_MS,
testName: 'office-hours-brain-writeback',
runId,
model: resolveEvalModel('capture'),
env: childEnv,
}),
validate: (result) => {
logCost('/office-hours (BRAIN WRITEBACK)', result);
expect(['success', 'error_max_turns']).toContain(result.exitReason);
// The headline assertion: agent actually called gbrain put on the
// expected slug.
if (!existsSync(callsLogPath)) {
throw new Error(
`No gbrain calls log at ${callsLogPath}. ` +
`Agent likely did NOT invoke gbrain at all. ` +
`Check that office-hours/SKILL.md in the workdir contains the gbrain put block.`,
);
}
const callsLog = readFileSync(callsLogPath, 'utf-8');
console.log('--- gbrain calls log ---');
console.log(callsLog);
console.log('--- end calls log ---');
expect(callsLog).toContain('gbrain put');
// Agent obedience: the slug should contain 'pixel-fund' somewhere
// (preferably under the office-hours/ prefix). The strict slug
// SHAPE (office-hours/<slug>) is already pinned by the resolver
// unit test (test/resolvers-gbrain-save-results.test.ts); this
// E2E proves the agent actually invokes gbrain put with the
// payload, not the resolver's literal output shape.
expect(callsLog).toMatch(/gbrain put .*pixel-fund/);
// Payload file exists. Agent may write to office-hours/pixel-fund.md
// (resolver-faithful) OR pixel-fund.md (agent dropped prefix); both
// are acceptable here because the YAML frontmatter is the real
// contract test. Search the payload tree for any *.md file that
// contains 'pixel-fund' in the path.
const findPayload = (dir: string): string | null => {
if (!existsSync(dir)) return null;
for (const entry of readdirSync(dir, { withFileTypes: true })) {
const full = join(dir, entry.name);
if (entry.isDirectory()) {
const nested = findPayload(full);
if (nested) return nested;
} else if (entry.name.includes('pixel-fund')) {
return full;
}
}
return null;
};
const payloadPath = findPayload(payloadDir);
if (!payloadPath) {
throw new Error(
`Agent called gbrain put but no payload file with 'pixel-fund' ` +
`in name was written to ${payloadDir}. Check the fake gbrain ` +
`--content parser for argv quoting issues.`,
);
}
const payload = readFileSync(payloadPath, 'utf-8');
expect(payload).toMatch(/^---\s*\n/);
expect(payload).toContain('title:');
expect(payload).toContain('tags:');
expect(payload.length).toBeGreaterThan(200);
// Entity stubs: agents are inconsistent about whether they use
// 'entities/<name>' (resolver doc) or 'entity/<name>' (singular).
// We accept either — the test asserts that AT LEAST ONE entity
// stub call exists, not the exact slug shape.
const entityCallMatches =
callsLog.match(/gbrain put entit(?:y|ies)\//g) || [];
if (entityCallMatches.length === 0) {
console.warn(
'No entity stub calls in gbrain calls log. Resolver instructs ' +
'entity extraction but it is best-effort.',
);
} else {
console.log(
`Entity stub calls observed: ${entityCallMatches.length}`,
);
}
},
});
},
CAPTURE_LONG_MS + OFFICE_HOURS_BUN_GRACE_MS,
);
},
);